Me too.

But I tell the students that almost all of the really interesting statistical procedures developed since WWII have been developed to address problems in the non-experimental fields. Almost all of the statistical procedures used in physics, or chemistry, or engineering, are pretty much similar to what Fisher and Jerzy Neyman (hallowed be his name) were doing in the 1930's. I work in the social sciences exactly because the data are crappier, we use observational rather than experimental data, and there's so much noise that we have to filter out to get to the signal. (As a not-so-aside, that's pretty much how I came up my approach to measuring aero and rolling drag -- I'm so used to crappy data in my regular day job that when I applied normal crappy data techniques to the relatively clean data from bicycle sensors the results turned out pretty well. It's also why a lot of engineers and physicists were pretty skeptical when I said I had done this. They think social scientists aren't real scientists and said, "that can't work because if it did, we would do it that way."). Linear regression was originally developed to investigate a social science problem; so was stochastic branching processes, which later was used by physicists to model nuclear fissioning and, um, the atomic bomb. Those were originally social science methods, taken over by the "hard" sciences. I'm very popular at parties.